Causality, determination and free will: towards an anscombean account of free action

Anscombe’s “Causality and Determination” is often cited in the contemporary free will debate, but rarely discussed in much detail. It’s main contribution, it is thought, is the defense of an alternative to deterministic causation, thus clearing the way for an incompatibilist analysis of free actions in terms of probabilistic causation. However, in this paper I will show that the contemporary probabilistic analysis of free action actually stands in direct conflict with Anscombe’s lecture. Instead, I will argue, its true value for incompatibilist accounts of free will lies in Anscombe’s thought that there are various fundamentally different kinds of causality. This variety, I argue, allows for a revised conception indeterminism and an understanding of free will as the manifestation of a particular sort of agent-causal power.

combe's idea that causation is not, by definition, a species of necessitation. (Franklin 2018, pp. 1-2). Given the wide recognition of the importance of C&D among contemporary incompatibilists, it is striking that the finer details of Anscombe's lecture are hardly ever discussed. In fact, as I will argue in part one of this paper, contemporary incompatibilists even seem to misconstrue some of the claims they deem so significant. One reason for the lack of explicit discussion of C&D is that it can hardly be said to be a defense of incompatibilism. The voiced commitment to incompatibilism rather figures as an addendum to a lecture mainly concerned with the broader notions of causality and determination. And so, in the eyes of contemporary incompatibilists, the work done in C&D is mainly of the ground clearing variety. Anscombe's merely negative role was to demolish the dilapidated conceptions of causation and determinism that stood in their way, and it was to them to erect new and full-fledged incompatibilist theories of free will.
The aim of this paper is to question this division of roles. In part one, I will attempt to show that one aspect of many contemporary incompatibilist theories of free will often attributed to Anscombe, i.e., their reliance on so called "indeterministic" or "probabilistic" causation, actually stands in direct conflict with C&D. This, however, does not mean that C&D is useless to contemporary incompatibilists. For, as I will argue in part two, Anscombe's thought, central to C&D, that there is more than one kind of causation gives rise to a substantive understanding of indeterminism and hints at a positive and promising incompatibilist understanding of human agency.

C&D and contemporary libertarianism
Libertarianism about free will consists of two claims. The first is the negative claim that free will is incompatible with determinism. The argument for this is simple and intuitive. Anscombe puts it as follows: "My actions are mostly physical movements; if these physical movements are physically predetermined by processes which I do not control, then my freedom is perfectly illusory" (C&D, p. 146). 1 The second claim is the positive claim that free will exists, which seems harder to defend. For, given the negative claim, this second claim implies at least two further theses: (1) that the world is in fact indeterministic, and (2) that free will is positively compatible with indeterminism. Many have expressed doubt about both theses. The common criticism of (1) is that we, as philosophers, cannot or should not make such strong claims about the empirical world. Surely, whether the world is deterministic or not is to be discovered by physicists, not philosophers. The problem with (2) is that indeterminism merely seems to inject randomness in the world. If a physical movement is not determined to happen, its occurrence must be a mere matter of chance or probability. And if there is anything we cannot control, it is the workings of chance. "Mere hap", it seems, "is the very last thing to be invoked as the physical correlate of 'man's ethical behaviour'", as Anscombe puts it (C&D,p. 145). So, although the libertarian's negative claim that determinism limits the control we have over our own actions might be intuitively plausible, the prospects for agential control in an indeterministic world seem even worse. These doubts made libertarianism into a highly unpopular position around the time Anscombe delivered C&D. 2 However, as contemporary libertarians are wont to emphasize, that paper itself presents an important step towards the current status of libertarianism as a respectable position-although perhaps still that of a minority-in the free will debate. In particular, Anscombe's paper is thought to help in defending the compatibility of freedom with indeterminism. Robert Kane, one of the main torchbearers for libertarianism in the eighties and nineties, credits Anscombe with the insight that "one can be said to "cause" or "produce" an outcome, even if that outcome is not inevitable or determined, given one's efforts" (Kane 1996, p. 55). Kane illustrates this with a convincing example: even though it might not be causally determined that a piece of radioactive material placed in an executive's desk drawer by a disgruntled nuclear facility employee will give the executive cancer, there still is no question that the employee can and should be held responsible if the executive does get cancer. As Kane writes: "it would be no adequate defense for the employee to plead that the outcome was a matter of chance and therefore he is not blameworthy" (Kane 1996, p. 55). It is thus a mistake, Kane argues, to conclude that a particular happening is outside of our control just because it is a matter of chance. Anscombe's argument that causation does not imply deterministic necessitation demonstrates, according to Kane, that there also is a different kind of causation, something he calls probabilistic causation, and his point is that "such causation is often good enough for ascriptions of power and responsibility" (Kane 1996, p. 55).
We can, however, doubt whether the disgruntled employee example is a good analogy for free human action. For Kane, when we act, the most we can do is make a serious effort of the will, and it is undetermined whether we achieve our intended result, just like it is undetermined whether the placement of radioactive material leads to illness. 3 An agent who makes an effort of the will thus has to wait and see whether it results in the action she actually wants to perform, and there is nothing she can really do to ensure that it will. Thus, although Kane's view establishes that we are responsible if we end up doing what we want, it seems that we still have no control over what we do-that remains up to chance. And hence, the compatibility of freedom with indeterminism remains questionable.
The contemporary libertarian, however, does not have to accept Kane's account of the role indeterminism plays in free action. According to Franklin, for instance, the problem with Kane's account is that Kane explicitly locates indeterminism after the choice: the agent does all she can to make something happen and then it is up to the world, and indeed beyond the agent's control, whether it will actually happen. But Franklin suggests that the problem of control is overcome if we locate the inde-2 As Peter van Inwagen recalls: Libertarians where "thin on the ground in the sixties and early seventies" (van Inwagen 2017, p. 3.) With Anscombe he also lists Carl Ginet, Roderick Chisholm,Peter Geach and Richard Taylor. 3 In actuality Kane's view is a little more complicated. He believes that most of our actions follow deterministic patterns but derive their freedom from a few indeterministic "character forming" actions which are based on torn decisions. These he analyses in terms of a dual effort of the will to make the one choice and the other. Although the results of these torn decisions are up to indeterministic chance, Kane contends that we are responsible whatever the outcome since either outcome will be probabilistically caused by an effort of our will. For the purposes of this paper, we can ignore these complications. terminism before the choice. 4 At first sight, it might seem that shifting the location of indeterminism in this way is not very helpful. It might no longer be a matter of chance what the agent will do after she has chosen, but if it remains a mere matter of chance what the agent will choose, how can that choice still be under her control? To answer that question, we have to understand that Franklin's thinking about control is informed by the causal theory of action. On that theory, sometimes referred to as the "standard story" (Smith, 2004, p. 165) due to its wide acceptance in analytic philosophy, an agent's exercise of control over her actions consists in her actions being appropriately caused by the right mental states. For instance, an agent does not control the movement of her lower leg if it is caused by a doctor who strikes her patellar ligament with a reflex hammer, but she does control the movement if it is caused by, say, her desire to kick a football. Now, how does this account of agential control help a libertarian like Franklin? Note that it would not be of any help if causation involved necessitation or exceptionless regularity, for that would imply that the mental states that bring about an intentional action would do so deterministically. But it is here, Franklin argues, that Anscombe's insights show their true libertarian worth. For she has shown that causation can also be probabilistic, thus opening up the possibility for an event-causal libertarianism. On Franklin's view, control just is a matter of probabilistic causation by mental states. An agent's decision to kick a ball manifests in a suitable mental state, e.g., her desire to kick it, causing her movement. The agent's freedom, i.e. her ability to not kick the ball, or to do something else, is reflected in the probability that her desire won't cause her to kick, or perhaps in the probability that another mental state will cause her to do some other incompatible thing. This puts us in a position to understand why Franklin thinks that the location of indeterminism in the causal chain leading up to the action matters. If it is the case that even after we make a decision to ϕ there still is a chance that we will not ϕ, then it seems that we have no control over our action. However, if the indeterminism is located before choice, then this merely means that chance determines which choice is made, but whatever choice is made, it will always be the agent's own mental states that cause the movement. And hence, because control is nothing other than causation by the appropriate mental states, whatever the agent ends up deciding, her action will always end up being under her control.
Franklin certainly is not the only, nor the first, contemporary libertarian that proposes to combine Anscombe's purported insight that causation can be probabilistic with Davidson's event-causal account of intentional action. 5 Not all contemporary libertarians, however, are on board with the event-causal analysis of action. So-called agent-causal libertarians believe that the best way to capture the intuition that free will requires the agent to have control over her actions is to think of the agent as "in a strict and literal sense an originator of her free decisions, an uncaused cause of them" (Clarke 2003, p. 134). But interestingly, many agent-causalists do think that the connection between the agent and her action is a probabilistic causal connection. 6 Tim O'Connor, for instance, argues that the agent-causal power is probabilistically structured by, among other things, the agent's reasons (O'Connor 2009, p. 205). And on Randolph Clarke's "integrated agent-causal account", free actions are at the same time probabilistically caused by the agent's reasons and agent-caused by the agent herself (Clarke 2003, pp. 133-150). 7 Looking at the contemporary free will debate, it might thus seem that the libertarian view on free will and the idea that we should understand the causation of free action in probabilistic terms have almost become synonymous, and more often than not Anscombe is championed to be responsible for this innovation. 8 Remarkably, however, in C&D, Anscombe explicitly rejects the idea that we should analyze human action in terms of probabilistic causation. She writes that, although it might well be that the "physically haphazard" is the "only physical correlate of human action": "The freedom, intentionalness and voluntariness are not to be analysed as the same thing as, or as produced by, the physical haphazard. Different sorts of pattern altogether are being spoken of when we mention them, from those involved in describing elementary processes of physical causality" (C&D p. 146).
Before we consider, in the part two, how we should think about the "different sorts of pattern" that, according to Anscombe, are involved in our understanding of free action, it will be instructive to consider what reasons she has to reject the analysis of free action in terms of probabilistic causation.
One reason might be that Anscombe considers it a mistake anyway to try to offer a causal analysis, probabilistic or otherwise, of the relation between reasons and actions. In her 'Practical Inference' (Anscombe 1989), when discussing the wellknown problem of deviance 9 for the causal theory of action, she writes: " [Davidson] speaks of the possibility of 'wrong' or 'freak' causal connexions. I say that any recognizable causal connexions would be 'wrong', and that he can do no more than postulate a 'right' causal connexion in the happy security that none such can be found. If a causal connexion were found we could always still ask: 'But was 6 An exception to this line of thought is Jonathan Lowe, who wants to understand agent-causation in terms of the exercise of non-causal powers (Lowe 2008). 7 In the remainder of part one, event-causal libertarianism will be the main target. But since the agentcausalists I mention also buy into probabilistic causation, I take many of the ensuing arguments to apply to their accounts as well. As we will see in part two, the account that I take Anscombe herself to be recommending is also of a broadly agent-causal variety. However, she would deny that the indeterministic agent causation involved in the production of free action is probabilistic in nature. I will say more about the similarities and differences between the Anscombean and the contemporary agent-causal accounts in part two. 8 The only exception to this picture is constituted by so-called non-causal libertarians who propose to understand human agency and free will in non-causal terms. These views go back to Carl Ginet (who is on van Inwagen's small list of libertarians before C&D in the sixties and early seventies). See e.g., McCann (2012), Palmer (2021). 9 This is the problem that a causal connection between a mental state and a corresponding movement does not guarantee the movement's intentional character, because the mental state might cause the agent to move accidentally, e.g. a mountain climber's desire to rid himself of the weight of the companion he is holding on to, might unnerve him to such an extent that he starts to tremble and accidentally lets go. the act done for the sake of the end and in view of the thing believed?'" (Anscombe 1989, pp. 378).
On the causal theory of action, there is nothing special about the kind of causality exhibited in intentional action. As Davidson is keen to emphasize: "rationalization is a species of ordinary causal explanation" (Davidson 1963, p. 685). 10 Hence, whether a mental state causes another mental state, a mouth to water, or the raising of an arm, the kind of causal connection is always the same. 11 But this implies, as Anscombe points out, that the mere existence of a causal connection between a mental state and a movement does not tell us anything about the justificatory relation between the two. The causal relation does not construe the one (the mental state) as a reason or motivation for the other (the movement). 12 Because of such criticisms, Anscombe is sometimes considered to be a non-causalist or anti-causalist about action. 13 But this is a mistake, for, as we will further discuss in part two, Anscombe does afford an important role to causation in her account of human agency. I therefore take Anscombe to suggest that the problem with the causal theory is not that any causal account of agency is wrong per se. It rather is that the event-causal theory of action conceives of causation in the wrong way. The problem she calls attention to -i.e., that a mere causal connection between a reason and a movement cannot tell us anything about whether the reason was the rational ground for the movement -seems to me to be a particular case of the more general worry, highlighted in C&D, that philosophical analyses of causation often fail to show in what way the cause is to be considered the source of the effect: [C]ausality consists in the derivativeness of an effect from its causes. This is the core, the common feature, of causality in its various kinds.
[…] Now analysis in terms of necessity or universality does not tell us of this derivedness of the effect; rather it forgets about that. For the necessity will be that of laws of nature; through it we shall be able to derive knowledge of the effect from knowledge of the cause, or vice versa, but that does not show us the cause as source of the effect (C&D, p. 136).
Once you adopt, as Davidson does, a neo-Humean analysis of causation in terms of natural laws that express regularities, you lose the conceptual resources to say anything about the way in which a cause is productive, or is the source, of its effect. For at the fundamental level described by these laws, causation is never more than simple contiguity. When it comes to action, neo-Humeans will thus not be able to say that the movement was produced in an intentional manner, e.g., that a certain desire caused an agent to move because the agent recognized that desire to be a good reason for her to move. The only way in which the causal theorist could try to include 10 It is, perhaps, this feature that makes the causal theory of action attractive to scientifically minded naturalists, and therefore contributed most to the status of the causal theory as "the standard story". 11 Anscombe actually protests against the thought that there is only one kind of causation throughout C&D. This will become crucial in part two of this paper. 12 Of course, proponents of the causal theory do not believe that a mere causal connection is sufficient either. If a movement is to be counted as intentional it should also be represented by the mental state that is its cause. However, as Horst (2015) points out, the causal theory necessarily construes the causal relation and the representational relation as logically independent, which makes causation blind to content and the possibility of causal deviance inescapable. 13 See Schwenkler (2019, Ch. 2). such a recognition of reason into her account, is by understanding the recognition as a mental event that itself must be a link in the causal chain leading up to intentional movement. This will not help, however, because even if the recognition-event plays a causal role in the production of an agent's movement, we can still wonder whether the agent moved out of her recognition of the reason, or whether that mental event caused it in some other (wayward) way. And it will obviously be of no help either to think about causal regularity as statistical rather than exceptionless, as the contemporary event-causal libertarian does. If we are to understand human action, then, it seems that we must insist that agency is a distinct mode of production. 14 Merely highlighting (probabilistic) causal connections will be insufficient.
A second reason Anscombe has to reject an analysis of free action in terms of probabilistic causation directly pertains to the thought (labeled "(2)" above) that freedom and indeterminism are incompatible since, as we saw, "'mere hap' is the very last thing to be invoked as the physical correlate of 'man's ethical behaviour'" (C&D, p. 145). Ascombe's response to this worry is twofold. On the one hand she takes it to correctly indicate that indeterminism by itself is insufficient for free will, because freedom "at least involves the power of acting according to an idea, and no such thing is ascribed to whatever is the subject (what would be the relevant subject?) of unpredetermination in indeterministic physics" (C&D, p. 145). But on the other hand the worry seems to assume that we need to analyze freedom in terms of probabilistic processes, purely because such processes might be the only physical correlate of human action. Hence the proper response to the worry, Anscombe seems to suggest, would be to offer an account of what it means to act according to an idea in non-probabilistic yet physical-indeterministic terms. Before we will take a closer look at this suggestion in part two, there are two further things worth highlighting.
The first is the sharp contrast between Anscombe's suggestion and the prevalence of probabilistic causation in contemporary libertarian accounts of free will. This again puts into question the contemporary libertarian's allusion to Anscombe as the originator of the thought that free acting might be a matter of probabilistic causation.
Secondly, we should consider whether Anscombe's agreement with the thought that an analysis of agency in probabilistic terms would turn actions into an uncontrollable matter of chance is outdated in the face of contemporary developments in the free will debate. For didn't Franklin show, as we discussed at the beginning of this section, that there is a consistent way to insist that agents do have control over their actions even if their choices are a matter of chance? As we have seen, he agrees with the causal action theory that an agent's exercise of control over her actions simply consists in her actions being appropriately caused by the right mental states, and maintains that it does not matter for this account that the appropriate causation is probabilistic in nature.
One reply to Franklin would be that the causal theory is to be rejected for the reasons we discussed above, i.e., that a mere causal connection cannot capture the justificatory relation between reason and movement. But even if we buy into Franklin's account of agential control, an issue remains. For, if Franklin is indeed correct that it does not matter whether the kind of causation involved in the production of action 14 We will come back to this in more detail in part two. is deterministic or not, then it follows that this account of control is also available to the compatibilist. In other words, once we accept the causal account of action, we thereby undermine the core negative claim of libertarianism, i.e., that determinism rules out control. 15 This result has not gone unrecognized by Franklin and other contemporary libertarians, who therefore see it as their task to explain why, even though some degree of agential control is available if the world is deterministic, indeterminism still allows for more control, and, moreover, that that extra bit of control is necessarily required for freedom of the will. Here is not the place to discuss and criticize Franklin's (or any other) answer to this so-called "problem of enhanced control" (Franklin 2011b) in detail. 16 But I do not think it surprising that responses to this problem often seem to resort to hollow-sounding rhetoric such as the claim that indeterminism is "worth wanting" because it affords "some independence from the past" (Mele 2006), or that indeterminism importantly grants agents "authorship" over their own live by affording them the "opportunity to direct their lives in more than one way" (Franklin 2011b). In reply to such remarks, I think the compatibilist would rightly wonder why independence from the past is worth wanting if agents can also control their actions without it, or why agents should be denied authorship, but not agential control, if determinism is true. 17 It thus seems that adopting the event-causal theory of action weakens rather than strengthens the libertarian position. Employing Anscombe's words, we can perhaps say that if the account of control or freedom you offer can also be employed to reconcile freedom with determinism, it makes "the alleged freedom quite unreal" (C&D, p. 146).
Contemporary libertarians often point to Anscombe and in particular to C&D as an inspiration for their account of free will. But, as I hope to have shown, the widespread acceptance of probabilistic accounts of the causation of action among contemporary libertarians actually is in conflict with her outlook. Moreover, I have argued that there indeed are good reasons (which I derived from Anscombe) for rejecting such probabilistic analyses, thus leaving contemporary libertarians with a deficient account of free will. It will therefore be worthwhile to investigate what different sort of account of human freedom we actually can distill from Anscombe's suggestions in C&D, and to highlight some of the advantages it might have over contemporary alternatives. That is the aim of the second part of this paper. 15 Indeed, Davidson himself exactly employed the causal theory in defense of his compatibilism. See Davidson (1973). 16 However, see (van Miltenburg & Ometto, 2019).

Indeterminism and the varieties of causation
In part one of this paper I have argued that C&D does not recommend an analysis of free action in terms of probabilistic causation. It might, however, be hard to see what the alternative to such an analysis would be, for it is often thought that probabilistic causation is the only alternative to deterministic causation. If that were correct, then there would obviously be no prospects for a non-probabilistic yet indeterministic account of free action. 18 Indeed, Anscombe herself is sometimes taken to argue against determinism on the basis of the possibility of probabilistic causation. Libertarians Stephen Mumford and Rani Anjum, for instance, write that: Anscombe's argument is designed to show that there is a coherent notion of probabilistic causation and that determinism cannot therefore be part of the notion of cause. (Mumford & Anjum, 2015, p.7, italics mine).
We also find this interpretation of C&D outside of the debate on free will. Davidson, for example, takes Anscombe's main criticism of his account of causationaccording to which every cause-effect relation instantiates a universal law-to be that it would "forbid the indeterministic laws of quantum physics" (Davidson 1995, p. 266). 19 These renditions of Anscombe's argument, however, can hardly be correct, given her explicit denial that indeterministic physics is logically required to make the deterministic picture doubtful (C&D p. 147), and her observation that even an unqualified acceptance of Newton's laws does not commit one to determinism (C&D p. 143). Before we say more about these admittedly rather cryptic remarks, allow me to speculate about the source of the mistaken interpretation. One possibility is that it is a symptom of the very (broadly neo-Humean) picture Anscombe wants to reject. 20 For, if one conceives of causation along classically Humean lines, as the instantiation of a law of nature (an exceptionless generality), then indeterministic causation is naturally thought of as instantiating a probabilistic analogue of such a law. On such a picture, then, there simply is no difference between arguing for indeterminism and arguing for probabilistic laws of nature. An argument for the one must at the same time be an argument for the other.
18 A (tacit) acceptance of the idea that probability is the only alternative to determinism perhaps also lies behind the thought that agential control is incompatible with indeterminism because it would render actions mere matters of chance. 19 Davidson even continues by suggesting that Anscombe might not have objected to his account after all, if only she had realized that indeterministic laws can also be universal and exceptionless in the sense that the probabilities they describe have no exceptions. To be fair, however, Davidson does show some awareness that his rendering of Anscombe's criticism is not entirely adequate by calling the possibility that she would no longer object "quite unlikely" (Davidson 1995, p. 266). It is also worth noting that right after this comment Davidson implicitly and, if I am right incorrectly, attributes to Anscombe the thought that free will should be understood in terms of probabilistic causation, when he claims that pace Anscombe, "the indeterminism of quantum physics cannot facilitate freedom of action" (Davidson 1995, p. 267). 20 I am not here suggesting that everyone who understands Anscombe's argument in the manner I am objecting to endorses a neo-Humeanism. Mumford and Anjum, for instance, in fact explicitly reject neo-Humeanism. My point is rather that the dominance of neo-Humeanism in analytical philosophy makes it hard to appreciate how radically different Anscombe's understanding of indeterminism is.
But, crucially, Anscombe wants to reject the idea that, at bottom, all causation consists in the regular co-occurrence of cause and effect. On her view there are very many different ways in which a cause can be productive of an effect. Each of these ways of production can be represented by a different causal concept. She lists scraping, pushing, carrying, squashing etc., and the more general concept of cause, she argues, can be added only to languages that already contain such particular causal concepts (C&D, p. 137). The view Anscombe recommends, I take it, is more or less in line with the metaphysical realism about powers that has (re)gained popularity over recent decades. 21 Causation, on that view, consists in the manifestation of the causal powers or dispositions of objects or substances: the hearth heats the wax which causes it to melt. Indeed, when Anscombe tentatively gives an example of a proposition that could be called a "law of nature" she offers a description of a power: "the flashpoint of such a substance is …" (C&D p. 138). 22 It is this alternative anti-Humean understanding of causation that, I believe, allows for a different understanding of indeterminism which ultimately creates the room for a non-probabilistic yet libertarian analysis of free action.
Before we consider that alternative Anscombean understanding of indeterminism, it is worth pointing out that the notion of a power should be familiar to anyone acquainted with the contemporary free will debate. For the, earlier mentioned, main incompatibilist rival to the event-causal libertarianism of e.g. Kane and Franklin is agent-causal libertarianism: the view that agents themselves produce their free actions by manifesting their agential powers. While it is true that the account I am taking Anscombe to recommend can ultimately be considered a variant of agent-causal libertarianism, there also are some clear differences between Anscombe's view and contemporary agent-causal theory. I will highlight these differences throughout this section, starting right now, as the first difference lies in the motivation for thinking about free action in terms of powers. Contemporary agent-causalists introduce powers in an attempt to accommodate the intuition that free agents themselves should be the ultimate source of their actions 23 , and not, like Anscombe, in order to develop an alternative non-probabilistic understanding of indeterminism. 24 In fact, as I have 21 See, e.g., (Bird 2010), (Greco et al. 2012). 22 It is often suggested that Anscombe is a singularist who believes causation cannot be explained by laws of nature at all. While I will not argue against this interpretation here, I do think that it betrays how much analytic philosophy is held captive by the neo-Humean picture. Being critical of the thought that causal relations are to be explained by Humean regularities, is simply equated with being critical of the thought that causal relations can be explained by laws of nature. See Ometto (2021) for an extensive and convincing criticism of the ascription of singularism to Anscombe. 23 Agent-causalism is often explicitly developed in direct response to an obstacle for event-causal theories knows as the "disappearing agent problem": the thought that the event-causal theory does not afford agents themselves any control over their actions because all the causal work is performed by events that take place in the agent rather than by the agent herself. See, e.g., (Lowe 2008, pp.159-161). 24 Ruth Groff (2019) is the exception here, for she does argue that a proper anti-Humean understanding of causation in terms of powers gives rise to a non-probabilistic understanding of agential powers she calls p-non-determinism. In as far as I understand Groff's view, p-non-determinism is the closest contemporary kin of the view I take Anscombe to recommend, and I believe Groff's paper and the current one could be viewed as different routes towards the same sort of position. Anjum (2013, 2015), perhaps are the second exception to my claim that contemporary agent-causalists do not primarily employ the notion of power to arrive at a different understanding of indeterministic causation. For on their account all mentioned above, some contemporary agent-causalists (Clarke 2003, O'Connor 2009) actively consider agent-causation to be itself probabilistic in nature.
To bring out how Anscombe's anti-Humean understanding of causation in terms of powers does give rise to a non-probabilistic understanding of indeterminism we will consider and expand on Anscombe's analogy between the laws of nature and the rules of a game (C&D, p. 143.) We can compare deterministic Humean laws with the rules of a simple card game where the players shuffle a deck of cards, divide it into two piles and then determine the winner by turning the cards over one by one to see who has the most red ones. Once the piles are made (once the initial conditions are set) this game is fully deterministic, and, importantly for our purposes, none of the players has any real control over what happens (as long as we hold fixed that they continue playing the game, that is). In a similar vein, I propose, we can compare statistical Humean laws with the rules of a simple dice game where the players take turns rolling a single die and add the resulting value to their totals until one of them wins by surpassing a score of fifty. This game is clearly indeterministic, and yet the players have no more control over what happens than they had in the card game. In the dice game we cannot from a given total game state derive all other states, as we can in the card game. We can only, at any point in the game, calculate the probabilities of what will happen. This difference, however, is irrelevant to the control the players can exhort. In both games the players only are mechanistically going through the motions that are required of them without having any true influence on their chances of winning.
This analogy, I think, nicely illustrates why the account of the probabilistic libertarians discussed in part one is unsatisfactory. They explicitly analyze the intentional production of our action as itself a matter of indeterministic causation subject to statistical regularities. 25 And while this analysis ensures that a decision process, like the game of dice, can have multiple different outcomes, it does nothing to show that the agents who engage in such decision processes have any more control than the players in the dice game.
The prospects for a more substantial understanding of agential control are much better, I believe, when we start thinking about the laws of nature in a different fashion: not as describing Humean regularities, but rather, as Anscombe proposes, much more akin to the rules of chess. These rules, among other things, describe the powers (active and passive) of the pieces on the board, but they do not predetermine the course of the game, nor do they say anything about the probabilities of what will happen. The rules rather set the antecedent possibilities of the game by ruling out certain happenings, such as the diagonal movement of a rook, and allowing others. Somepowers always only have a tendency to produce their result, and hence all powers in a sense are indeterministic. Discussing Mumford and Anjum's conception of indeterminism and its implications for free will here would take us too far afield, however, see (van Miltenburg & Ometto 2020) for a critical engagement with their view. 25 Indeed, many contemporary libertarians seem to believe that the indeterminism they see on the level of human decision making can ultimately be traced back to the indeterminism we find at the neuronal or the micro-physical level. Kane, for instance, hypothesizes that efforts of the will "are complex chaotic processes in the brain, involving neural networks that are globally sensitive to quantum indeterminacies at the neuronal level" (Kane 1996, p.130). And Franklin argues for the naturalistic plausibility of libertarianism on the basis of the observations concerning the indeterministic workings of the brain (Franklin 2013). times the position might be such that, given the rules, only one move is possible. But forced moves are rare and at all other times the players themselves can control which (proximate) possibility gets actualized. If the laws of nature are like the rules of chess, then there is room for a third option: what happens would not have to be either completely fixed by deterministic laws or randomized by indeterministic statistical laws. 26 Instead, there could be what Arthur Prior (1962) has called a "limited indeterminism", where the powers described by the laws of nature determine what is possible at every time and place, without determining, or determining the probability of, what will actually happen. 27 We are now in a position to understand the alternative Anscombe has in mind when she writes that probabilistic physics is not required to put indeterminism into question. For what she believes "is fanciful" is that the laws, whether they are deterministic or probabilistic, would totally cover "every motion that happens" (C&D, p. 147). On a limited indeterminism, by contrast, the laws leave open multiple possible motions. And just like a chess player does not break the rules by making one among a multitude of different possible moves, animals, for instance, can "run about the world in all sorts of paths" without breaking any laws (C&D, p. 143). 28 It is the openness of the future afforded by limited indeterminism that, I think, allows for agential control. Because the laws of nature do not settle everything, there is room for us, free human agents to partially determine what happens. But limited indeterminism by itself of course remains insufficient for free action. A radium-226 atom that manifests its power to decay also realizes the previously open possibility of emitting an alpha particle, and yet it is in no sense free. In order to understand free action, it thus seems that we must explain what differentiates our agential powers from other non-deterministic powers.
This challenge is not uniquely faced by the Anscombean account. In fact, it is particularly pressing for contemporary agent-causalists, who are generally very emphatic about the ubiquity of causal powers throughout nature. 29 The reason for 26 This, of course, is not meant to imply that there is no place for deterministic or probabilistic laws whatsoever. Laws describe powers and, as Ometto (2021) explains, we can distinguish between deterministic and indeterministic powers. Some of the latter, such as radium's power to decay, are probabilistic. 27 Prior argues, in line with the reading of Anscombe I present here, that a metaphysics of substances and powers is required to spell out an indeterminism that does not make everything that happens contingent and random. 28 The idea that the laws need to cover everything that happens might also be rejected by those neohumeans who think about the totality of laws as "the best system" of regularities (Lewis 1973). The best system always strikes a balance between descriptive strength and other theoretical virtues such as simplicity. It might therefore be the case that some happenings are not covered by the best system of regularities when a system that would include them would be worse (e.g., because it would only be slightly descriptively stronger at the cost of being much more complex). But a crucial difference remains between the Humean picture and the view I take Anscombe to be recommending. For if one combines the best systems idea with the Davidsonian thesis that every causal relation must instantiate a natural law, it follows that the happenings which are not covered by the best system cannot be causally explained precisely because they fall outside the purview of natural law. On a limited indeterminism, by contrast, we can still think of the realization of one of the open options as causal production i.e., as the manifestation of a power. This is crucial for our purposes because freedom, as Anscombe claims and we will discuss shortly, "at least involves the power of acting according to an idea" (C&D, p. 146). 29 See, e.g., (Clarke 2003, ch. 10), (Lowe 2008, ch. 8). their emphasis is that they want to distinguish themselves from earlier agent-causalists (e.g., Chisholm 1966) who held that human actions are the only events in nature that are agent-caused and were therefore accused of 'obscure and panicky metaphysics' (Strawson 1962, p. 27). Contemporary agent-causalism overcomes this objection by understanding agent-causation (or better, substance-causation) as a species of power manifestation that occurs throughout nature, rather than as a uniquely human, unnatural, and mysterious phenomenon. But it should be noted that this insistence on the ubiquity of substance-causation at the same time undermines any appeal to substance-causation as the defining feature of free will. If powers are everywhere, then what makes the exercise of our agential powers uniquely free? My purpose is not to evaluate the ways in which contemporary agent-causalists answer this question. 30 Instead, I mention it in order to draw our attention to a second difference between contemporary agent-causalism and the Anscombean account I am developing. For contemporary agent-causalists, in stressing the ubiquity of powers, can make it seem like there is only one kind of causation: power manifestation. And thus, agent causalists are prone to the same mistake Anscombe ascribes to neo-Humeans: the inability to recognize that there are many different varieties of causal production. 31 There is, however, nothing contradictory about holding both that all causation is a matter of power manifestation and that power-manifestation is a genus that allows for different species of causal production. 32 It is this second claim, which is emphasized by Anscombe, that, I believe, can help us understand why only agential powers give rise to freedom. For if causation is not of a single kind, then not all settling of open possibilities needs to be blind and random as is the case with radioactive decay.
The idea that there are different varieties of causation exemplified by different kinds of power, can be naturally extended to the thought that these powers might be operative at different levels of reality. There might be (micro-) physical powers such as charge, spin or mass, but there also might be chemical powers, such as the power to dissolve or melt and biological powers such as the powers of growth, perception and self-movement. Note that the idea that there are genuinely causal powers that operate at different levels, gives rise to a second sense in which the laws can fail to cover everything that happens. Not only can it be the case that the laws of a single level fail to cover every motion at that level, they might also fail to cover what happens on a different (higher) level. 33 This, I take it, is one of the points Anscombe is trying to make with the example of the Coca-Cola box: 30 One attempt is to say that agential powers are unique in that they are "two-way powers" (e.g. Steward 2012): our power to act also is a power not to act. But at least prima facie this does not help very much, for why can't we say that a radium atom also possesses a two-way power-the power to decay or not to decay? Thus it seems that if this suggestion is to work, we need an account of the unique way in which agential powers are two-way. See (van Miltenburg & Ometto 2020) for an attempted specification of such an account. 31 Especially Lowe's (2008) account of substance causation seem eerily similar to the standard neo-humean conception of event causation. Event causation is an extensional relation between two events, substance causation simply swaps out the first event for a substance. 32 Again Groff (2019) is exceptional compared to other contemporary agent-causalists in stressing exactly this point. 33 In "The Causation of Action" (1983), Anscombe is more explicit about the existence of multiple levels of reality and about why we should think about higher level causes as doing genuine causal work, rather than "Suppose that we have a large glass box full of millions of extremely minute coloured particles, and the box is constantly shaken. Study of the box and particles leads to statistical laws, including laws for the random generation of small unit patches of uniform colour. Now the box is remarkable for also presenting the following phenomenon: the word "Coca-Cola" formed like a mosaic, can always be read when one looks at one of the sides. It is not always the same shape in the formation of its letters, not always the same size or in the same position, it varies in its colours; but there it always is. It is not at all clear that those statistical laws concerning the random motion of the particles and their formation of small unit patches of colour would have to be supposed violated by the operation of a cause for this phenomenon which did not derive it from the statistical laws" (C&D, p. 146).
What the example shows is that even if every motion at the micro-physical level is governed by (statistical) laws, this still does not rule out that there also are higher level phenomena that have higher level causes. Something that is left open by the laws of a lower level (such as that the word "Coca-Cola" must always appear) might thus be settled by a cause at a higher level, without this interfering with the statistical laws. 34 Now, how does this help us to understand free action? Consider the difference between the way in which a radium atom settles what happens by decaying, and the way in which the chess players determine the course of the game. On the microphysical level, which open possibility (decay or no decay) gets realized might indeed be nothing more than a mere matter of chance. No other way of settling open possibilities seems to be available. 35 But the players in the chess game do not need to determine their moves by rolling a dice or flipping a coin. They rather decide what happens based on what they deem to be the best move. The players seem to possess a higher level power, "the power of acting according to an idea" and it is this power, Anscombe tells us, that is required for freedom (in addition to physical indeterminism; C&D, p. 146). Anscombe thus suggests, it seems to me, that among the varieties of causal powers, we find what we could call rational powers. And moreover, that it is via these rational powers that we exert out freedom to settle which hitherto open possibility gets actualized.
It is, however, no easy feat to explain exactly how a power can be rational. For how are we to understand the relation between an agent's ideas (her motivations, reasons, desires) and her power to act in accordance with them? Perhaps the difficulty of this question is reflected by the fact that different contemporary agent-causalists answer it in widely divergent ways. Clarke (2003), for instance, argues that when an agent exercises her causal power to bring about her movement, this movement is at the same time also indeterministically caused by the agent's reasons in the ordinary event-causal manner. Lowe (2008) disagrees that agent-causalism should be supplemented with an event-causal account of acting for a reason. He rather thinks that we as affording abbreviated ways of speaking about a fundamental level. See Mulder (2021) for an in-depth exploration of the idea of a multiplicity of levels that gives rise to multiple varieties of indeterminism. 34 For a more technical discussion of the workings of such a box, see Müller (2022). exercise our agential-powers "in the light of reason", which in turn can be understood in terms of the exercise of our non-causal power of willing. O'Connor (2009), offers yet a different account and argues that reasons "probabilistically structure" our agential powers. On his view, reasons do have a causal influence on the agent's power to form the intention to act by altering the propensity of her exercising that power. This is not the place to assess the merits and flaws of these individual proposals. 36 What I do want to note here is that all of them depend quite heavily on ideas springing from the very theories of action-(probabilistic) event-causal action theory and non-causal action theory-to which agent-causalism was supposed to be the alternative. In the light of this, it would be interesting to see whether it is possible to also develop the thought that there are rational powers in more purely agent-causal terms.
This, I think, could be a motivation for contemporary agent-causalists to take a closer look at Anscombe's ideas. For we should note that Anscombe's suggestion in C&D that our freedom resides in our uniquely rational power to act according to an idea cannot be read in isolation. It rather is part of the inaugural lecture of a thinker who devoted a large part of her philosophical life to understanding intentional agency. Of course, an extensive discussion of Anscombe's philosophy of action would take us far beyond the scope of this paper, but I nevertheless would like to briefly highlight some features of action she identifies in order to indicate how an Anscombean understanding of the power to act might benefit libertarianism about free will.
Anscombe observes that actions are not intentional tout court but only "under a description" (Anscombe 1957, p. 12). This means that a particular movement might be intentional when it is described as "pinning the queen" but not when it is described as "shifting the center of gravity of the chess-set by 3 mm", for the second description indicates an effect the agent does not at all intend to produce and in fact might not even be aware of. 37 To describe a movement as an intentional action thus is to offer a description of the movement that is in accordance with the idea the agent has in performing that movement.
This thought that actions are only intentional under a description is sometimes referred to as the "Anscombe-Davidson thesis", because this idea also forms a staple of Davidson's seminal event-causal account of action. 38 However, I think that the way in which both philosophers give substance to this thought is importantly different. For Davidson, the fact that only certain descriptions of an agent's movements are in accordance with the agent's thought is irrelevant to the causation of that movement. This is because, as we have touched upon in part one, Davidson conceives of the relation between cause and effect as mere extensional regularity that holds between two events or objects however they are described. So, although Davidson holds that we can only describe a movement as intentional when it is both causally related to, as well as in accordance with the agent's thought, he thinks that these relations of accordance and causation themselves are independent. 36 However, see (van Miltenburg & Ometto, 2019) for some critical discussion. 37 Although it might also be a foreseen side-effect, see (van Miltenburg, 2011). 38 I thank an anonymous referee for encouraging me to clearly differentiate Anscombe's and Davidson's appeal to the doctrine that actions are intentional under a description.
Anscombe, I believe, would disagree with this. For note that one way of understanding the thought, central to C&D, that there are different kinds of causation, is to say that causation itself happens under a description. Recall that causation, for Anscombe, does not consist in mere extensional regularity, but rather specifies a particular mode of production. Now such a concrete form of production is only ascribable to an object when it is understood in a substantive way, i.e., under a description. It does not make sense, for example, to say of a particular region of space time that it melts, for only material substances melt. Nor does it make sense to say of a mere lump of matter that it perceives anything, as only certain organisms possess the (biological) power of perception. Anscombe's suggestion that we have a causal power to act in accordance with an idea thus implies that there is a particular intentional or rational mode of causal production. Her idea, it seems to me, is that we can produce actions, i.e. the pinning of the queen, under that description. Contrast this again with Davidson's event-causal analysis of action. In part one we discussed the worry that the existence of an extensional causal relation between an agent's movements and her reasons does not establish that the agent acted on the grounds of these reasons, giving rise to the possibility of causal deviance. We suggested that this problem prevails because the neo-Humean lacks the conceptual resources to say anything about the way in which the cause produces the effect. 39 Now we can see how an Anscombean account is different, for on such an account there simply is no difference between the agent acting on the grounds of her intention and the agent exercising her causal power to act (and therefore no room for deviance). To act intentionally is to causally produce a movement where the mode of causality is specifically rational. 40 In the wake of the recent resurgence of interest in Anscombe's philosophy and in particular in her theory of action, there have been several interesting attempts to develop this notion of rational causation along Anscombean lines. 41 One core idea is that rational causation consists in the exercise of the capacity for practical knowledge. When an agent, for instance, is cutting onions, she knows without observation that she is doing so and why she is doing it: she is making risotto. So, while the mere act of cutting onions is compatible with the production of all kinds of dishes, it is the agent's practical knowledge that determines under which description the agent's movements actually fall: this cutting of onions is part of making risotto. So not only is it the case that the agent's reasons for making these particular cutting movements are known to her in the exercise of her power to act, it is this knowledge that specifies that exercise: the knowledge makes the exercise into risotto cooking.
Rational powers might thus be different from powers we find elsewhere in nature in that their manifestations are specified by our thoughts. Whereas e.g., the power of iron to melt needs no specification and can only be exercised in one way (the iron 39 Indeed, the problem of deviance seems to prevail precisely because of the independence of the relations of causation and accordance to which Davidson subscribes. See Horst (2015). 40 This, however, should not be taken to imply that nothing more can be said about this particular mode of rational causal production, i.e., about how recognizing grounds and causing can be one-but only that such an account will not be reductive in the sense that it will oppose Davidson's claim that "rationalization is a species of ordinary causal explanation" (1963, p. 685). 41 See, e.g., Rödl 2007, Marcus 2012 simply gets hot and thus melts), our power to act can manifest in all the different kinds of actions we know how to perform. 42 Of course, I would need to say much more than I can say here to further develop the Anscombean account of rational causation in terms of practical knowledge. But what I do want to urge here is that the project of understanding the rational power to act should be squarely on the libertarian agenda. 43 For if we have a genuinely rational-causal power to act, then that gives rise to its own rational level of description. And the phenomena (the actions) that are brought about at that level should not be thought to violate the laws of other (lower) levels of description. Just as the formation of the word "Coca-Cola" is realizable by many different configurations of the minute particles in the box, so the action of "pinning the queen" is realizable by an infinite amount of slightly different bodily movements. When we describe such a bodily movement in lower-level terms, e.g., by specifying the movement of every micro-particle that makes up an arm, or by detailing which muscles contract to what extent on the signal of which afferent nerves, we fail to offer a description under which that movement is intentional. The agent is not even aware of the particular physical instantiation of her movement, let alone that she intends to produce just that instantiation; she merely wants to pin the queen. And just like a causal explanation of the movement of the individual particles would fail to capture the cause of the higher level phenomenon that Coca-Cola is always readable, lower level causal explanations of the particular bodily movements we make fail to causally explain the action under its intentional description. 44 This, it seems to me, is what Anscombe has in mind when she writes that "different sorts of patterns altogether" (different from the probabilistic patterns described in microphysics) should be spoken of once we consider the power to act in accordance with an idea (C&D p. 146.) Our rational power to act thus is the power to actualize, on grounds of our own reasons, one of the possibilities left open by non-rational nature. Such a power, I think, is deserving of the name "free will". 42 This might be one way to give substance to the thought we came across in footnote 30 that rational powers must not be one-way powers, but instead might have multiply different manifestations. See (van Miltenburg & Ometto 2020) for an elaboration of this idea. 43 Despite the recent resurgence of an interest in practical knowledge and Anscombe's philosophy in general it is rarely discussed in the context of vindicating a libertarian account of free will. I believe that this in part is due to the way in which contemporary libertarians frame their own mission, i.e. as the attempt to figure out what "transform[s] a mere action into a free action" (Franklin 2011a, p. 203). For Anscombe, the question what makes a movement an intentional action cannot be so easily separated from the question what makes it free. Anscombe might thus be seen as a proponent of what Helen Steward (2012) has recently called "agency incompatibilism": the view that agential control itself is incompatible with libertarianism. Also see (van Miltenburg & Ometto, 2019) for an argument that the framing of the free will debate as isolated from the philosophy of action skews that debate in favor of compatibilism. 44 This, I take it, is why Anscombe (1983, p. 101) argues that a physiological description of a bodily movement and explanations of the movement in terms of the reasons for which it was performed should not be thought of as rival causal accounts.

Concluding remarks
In part one of this paper, we discussed the widespread idea that the production of action is a matter of probabilistic causation. Despite the fact that C&D is often cited as the source of the thought that causation can be probabilistic in nature, I have argued that the analysis of free will in terms of probabilistic causation stands in direct conflict with what C&D has to say about that topic. In part two of this paper, I have suggested that another idea developed in C&D, i.e., the thought that there are different varieties of causation, offers a lot more promise to the contemporary libertarian and have contrasted the resultant Anscombean approach with other related contemporary agent-causal approaches. I have briefly discussed two merits of the Anscombean approach. First of all, the multiplicity of causal powers gives rise to a limited indeterminism on which the laws of nature delineate the future possibilities while leaving room for us, human agents, to act and thereby determine which possibility becomes actualized. Secondly, if we emphasize the various different kinds of causation, rather than the ubiquity of powers, there might be room for causation of a specifically rational kind. Such a power to act would vindicate an idea that is at the core of our notion of free will, i.e., that we can control what happens on the basis of our own reasons.
What I have offered in this paper is of course but a mere sketch of the lines along which an Anscombean libertarianism could be advanced. I have not developed such a libertarianism in any detail, nor have I defended it from possible criticisms. Clearly, much more work needs to be done. But I believe that this work is worthy of doing because the Anscombean account equips us with clear answers to the two main criticisms leveled against incompatibilist accounts of free will we touched upon at the very beginning of part one. The first objection was that whether the world is indeterministic is something physics, not philosophy, should discover. But the view I take Anscombe to recommend is resistant to that thought. For physics only is the investigation of the powers and processes that are operative at the (micro-) physical level, and the world might be indeterministic precisely because the laws on that level do not cover every phenomenon. Perhaps the objector might want to respond that higher level phenomena ultimately must receive micro-physical explanations as well. Such a response, however, undermines the original appeal to science, for such reductionism is itself a presupposition that is in need of philosophical, rather than empirical, justification. 45 The second objection was that indeterminism would merely inject randomness into the world and, since we cannot control matters of chance, would therefore thwart rather than enable human freedom. As we have seen, the Anscombean response to this objection is twofold. On the one hand, contrary to the belief of most contemporary libertarians, we can agree with the objector that freedom would indeed be unavailable if our actions where mere matters of chance. But on the other, we can simply deny the identification of indeterminism with the injection of randomness. As the picture of limited indeterminism shows, indeterminism might simply consist in the fact that the laws leave open multiple (perhaps higher level) possibilities-possibilities that we ourselves can actualize. 45 See Mulder 2021 for a discussion of the limits of such a reductionism.
One of the difficulties of philosophy is that it can be so easy to lose oneself in the construction of complex theories and discussions about the minute differences between them-so much so that we sometimes almost seem to forget about the original concept or phenomenon we were trying to understand. The contemporary debate about free will is, I think, not exempt from this phenomenon. We find many intricate accounts that try to secure some notion of freedom that remains available even after admitting that our actions might be predetermined by natural law or governed by mere chance. However, the original thought that human beings possess free will, it seems to me, is that it neither is the case that the deck of what happens is already stacked nor that our actions are subject to the mere roll of a die. It rather is the thought that we ourselves can have partial control over what happens. If the interpretation I have levied in this paper is apt, then C&D offers a picture of the world, its laws of nature, and the causally powerful beings that inhabit it, that creates room for such self-determination.